Xiao Li

xl2710 [at] princeton.edu

Hi! My name is Xiao. I'm a CS PhD student at Princeton University in the vision/graphics area. I'm interested in developing novel physical AI systems for applications in imaging, computing, robotics, and beyond by combining system design with computational methods or machine learning.

Recently, I have been focused on applying machine learning to the inverse design of metalens cameras for photography and optical computing. Before coming to Princeton, I worked at Magic Leap and contributed to the imaging/display optics and the SLAM-based head pose tracking system of AR headsets. I also conducted research in Terahertz photonics at UCLA advised by Prof. Mona Jarrahi.

Currently, I'm on leave and working at Meta Reality Labs.

---

Publications

“*” stands for equal contribution.

Spatially varying nanophotonic neural networks
Kaixuan Wei , Xiao Li , Johannes Froech , Praneeth Chakravarthula, James Whitehead, Ethan Tseng, Arka Majumdar, Felix Heide
📚 Science Advances

• Summary: Demonstrated a compact metalens array camera that performs neural network convolution during the image capture. This optical computing prototype performs >99% FLOPS with photons and outperforms AlexNet on CIFAR-10 blind test classification accuracy for the first time.

Thin On-Sensor Nanophotonic Array Cameras
Praneeth Chakravarthula, Jipeng Sun, Xiao Li , Chenyang Lei, Gene Chou, Mario Bijelic, Johannes Froesch, Arka Majumdar, Felix Heide
📚 SIGGRAPH ASIA 2023

• Summary: Demonstrated a broadband metalens array camera with learned probabilistic deconvolution to improve the image quality.

Seeing Through Obstructions with Diffractive Cloaking
Zheng Shi, Yuval Bahat, Seung-Hwan Baek, Qiang Fu, Hadi Amata, Xiao Li , Praneeth Chakravarthula, Wolfgang Heidrich, Felix Heide
📚 SIGGRAPH 2022

• Summary: Proposed a computational monocular camera that cloaks unwanted obstructions by jointly designing a diffractive optical element at the aperture plane and a feature-based deep learning reconstruction network to recover the unobstructed image.

Tunable terahertz wave generation through a bimodal laser diode and plasmonic photomixer
S.-H. Yang, R. Watts, X. Li , N. Wang, V. Cojocaru, J. O’Gorman, L. P. Barry, and M. Jarrahi
📚 Optics Express

• Summary: Demonstrated a compact broadband terahertz source based on difference frequency generation by pumping a plasmonic photomixer with the dual frequencies from a two-section digital distributed feedback laser diode.

Patents

• Eyeball Camera for Display System Calibration, US 2021/0014477, WO/2021/011420, CN114128256, EP3997867
• Waveguides with High Index Materials and Methods of Fabrication thereof, US 2022/0128817, WO/2020/185954
• Systems and Methods for External Light Management, WO2020/023266, US 2021/0231951, CN113811803, EP3938824

Non-work stuff

• When I'm not playing with cameras and pixels in the lab, I like taking photos, with street photography being my favorite. Here's selected photos that I took with an iPhone.